yolo

yolo TASK MODE [arg=value...]

yolo is the CLI for Ultralytics YOLO, a state-of-the-art computer vision framework. It provides commands for object detection, instance segmentation, image classification, pose estimation, and oriented bounding box detection from the terminal.The tool supports a complete workflow: train builds models from datasets, val evaluates model accuracy, predict runs inference on images or video, export converts models to deployment formats like ONNX and TensorRT, track performs multi-object tracking on video streams, and benchmark tests model performance across formats.Each command accepts an optional task type (detect, segment, classify, pose, obb) and a required mode. Arguments are passed as key=value pairs. Pre-trained models can be used directly for inference or fine-tuned on custom datasets. GPU acceleration is supported through PyTorch.

## Tasksdetect

Object detection.

Instance segmentation.

Image classification.

Pose estimation.

Oriented bounding box detection.

Train a model on a dataset.

Validate model accuracy.

Run inference on images, video, or streams.

Convert model to deployment formats (ONNX, TensorRT, CoreML, etc.).

Multi-object tracking on video.

Benchmark model speed and accuracy across export formats.

Model file path (e.g., yolo11n.pt).

Dataset configuration YAML file.

Input source: image, video, directory, URL, or webcam (0).

Number of training epochs.

Input image size (default: 640).

Batch size.

Device: GPU id (0, 0,1) or cpu.

Export format: onnx, engine, coreml, tflite, etc.

Confidence threshold for predictions.

Requires Python and PyTorch. GPU recommended for training. Pre-trained model weights are downloaded automatically on first use. Use `yolo cfg` to view all available configuration arguments.

python(1), pip(1), nvidia-smi(1)